Build and deploy your research and AI workloads with ease and efficiency
Fuzzball takes a researcher-first approach with intuitive tools and a built-in Web UI. Anyone on your team can build and deploy workflows from day 1 without infrastructure expertise. For administrators, it provides unified visibility, policy controls, and seamless integration with existing schedulers.
Observable data management
Fuzzball service endpoints provide interactive access to your workflows
Unify batch, interactive, and AI workloads by embedding long-running services directly into your workflow definitions.
Service endpoints in Fuzzball allow you to inspect simulations in real time and combine AI training and inference into a single, cohesive pipeline. They give you access to live results via Jupyter notebooks without requiring you to manage separate sessions or maintain manual connections.
External services
Run Jupyter notebooks and virtual desktops in-browser. Connect debugging tools to live workflow data without the friction of SSH tunneling.
Internal services
Launch databases and API servers instantly. Fuzzball handles the setup, scaling, and teardown automatically as your jobs run.
Are you currently using PBS or Slurm?
Why Fuzzball
For decades, high performance computing forced researchers to also become computer scientists. They spent more time mastering complex IT systems than advancing their actual research.
The CIQ team rejected that trade-off. Building on the container revolution that began a decade ago, Fuzzball utilizes orchestration to manage complex hybrid clusters. By abstracting infrastructure, it lets experts focus on their research, ensuring HPC serves the user rather than demanding their attention.
Focus
Focus on your research and not the infrastructure needed to run your jobs.
Freedom
Deploy your compute intense work anywhere without special systems expertise.
Modern
Employ modern infrastructure so you can scale and deploy more efficiently.
Why Fuzzball for your teams
Fuzzball combines the decades of performance and optimization research from traditional HPC, and combines that into the next generation solution that is “right” for your teams.
Researchers: Gain autonomy
Deploy workflows through a Web UI; no tickets, no command-line expertise required.
Administrators: Get control
Unified visibility across batch jobs, containers, and interactive services with policy-driven resource allocation.
Organizations: Improve productivity
Faster time-to-insight with infrastructure that scales without adding operational overhead.
Why Fuzzball for your use case
Fuzzball is applicable across a wide range of use cases but some of the more interesting include the following:
Traditional HPC + AI
Run molecular dynamics with machine learning training without infrastructure silos.
Bioinformatics pipelines
Execute Nextflow workflows with automatic resource scaling and data management.
Interactive analysis
Submit batch jobs through Slurm; access results through Jupyter notebooks.
Hybrid cloud bursting
Develop locally, burst to cloud for scale, deploy to production and use the same workflow.
Fuzzball works the way you work
Fuzzball fits your software development supply chain and helps you use the tools and infrastructure you require.
Works with your applications
Fuzzball comes with integrations for the tools you already use with integrations and templates so anyone can use them.
Works with your tools
Integrate Fuzzball into your CI/CD pipelines with API-driven automation for streamlined job and resource deployment.
Works on your infrastructure
Fuzzball comes with integrations for GPUs, TPUs, and architectures with no custom drivers or connectors needed.
Fuzzball works with your GPUs
Built on the foundation of CIQ's NVIDIA CUDA partnership, Fuzzball delivers simplified orchestration of GPU-accelerated HPC workloads through pre-validated container environments.
Manage your most demanding HPC and AI workloads with Fuzzball's orchestration intelligence powered by CUDA-optimized Rocky Linux from CIQ infrastructure.
Fuzzball in action: Build and deploy your workflows with ease
Fuzzball's Web UI is designed for usability and standardization. Anyone can build, manage, and reuse workflows with just a few clicks:
1: Create a new workflow
Import from a YAML file, copy a previous workflow, or start from scratch with a built-in template.
2: Define your jobs
Parameterize and customize every step of your workflow to match the needs of your jobs.
3: Customize your environment
Specify infrastructure requirements, from instance sizing to storage volumes, ensuring your workflow runs efficiently.
4: Run, share, and reuse
Fuzzball workflows are modular and portable. You can run them on any Fuzzball clusters. Create templates, save your workflows, and share them with your team.
The data and results from one workflow can be the input to another workflow. Chain, sequence, and customize your workloads all within Fuzzball.
What's new in Fuzzball
PBS/Slurm integration
Fuzzball now integrates with PBS and Slurm as backend provisioners. Keep your scheduler.
Workflow catalog
Get started in minutes with ready-to-run templates for HPC, AI/ML, and interactive environments.
Fuzzball resources
Documentation
Our docs are online so you can explore deeper concepts and get a sense of how it works.
Tech session
In this 45-minute session, we walk through the basics of Fuzzball and look at the future of modern performance-intensive platforms.